Jin Hsueh @chenian

@DavidSacks SaaS isn’t dead. The interface is dying. The system of record isn’t.

Shipping a product publicly takes more courage than most engineers admit. Because the moment you ship, everyone gets to inspect your decisions. I recently saw a subtitle tool built by a creator in Taiwan. My first reaction as an engineer wasn’t: “What stack did they use?” It was: “Damn. How long did it take to make the UI, UX, and entire experience feel this polished?” Then I saw the comments. Architecture is wrong. Security could be better. This shouldn’t be built that way. Those criticisms can be valid. But I think engineers sometimes optimize for the wrong scoreboard. A product is not a technical exam. Someone took an idea from 0 → 1, found a painful problem, built something people actually use, and convinced some of them to pay. That deserves respect. Engineering matters. But so do: Finding the right problem. Designing the experience. Getting users to stay. Getting users to pay. The best architecture in the world cannot rescue a product nobody wants. I’m increasingly convinced that shipping teaches you things code review never can. When you see an imperfect product getting real traction, do you first see the flaws—or the thing they got right? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

@R01keshh The bottleneck moved from shipping code to finding truth.

Our user didn’t abandon onboarding because the AI was hard to train. They abandoned it because they had to design the business workflow themselves. Customers had already trained CAAC Go on its accounting and tax content. Then they stopped. They wanted the AI to answer basic questions—but not answer everything. The real goal was to build trust, guide visitors toward a consultation, collect lead details, and hand off to a human. They knew the business outcome. They didn’t know how to translate it into AI behavior: what to answer, where to stop, which questions to ask, and when to escalate. So “train the AI” became a strategy project. And strategy projects lose priority when real work arrives. This echoed feedback from a US product designer: our activation flow showed users what they could configure, but not what outcome they could create, how to create it, or why they should trust the result. A blank AI configuration screen isn’t flexibility. It’s work. Our next activation challenge isn’t simplifying settings. It’s turning business outcomes into guided recipes. How much should an AI product configure for users before flexibility becomes a limitation? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

@jasonlk Data can be migrated. Context has to be accumulated.

I learned what Shopify does well. Then I realized the goal isn’t to beat Shopify on features. It’s to make sure a Shopify merchant never tries CAAC Go and thinks: “Why is this harder?” The biggest lesson: AI products lose when they expose their infrastructure. A merchant shouldn’t need to understand RAG, embeddings, knowledge bases, or system prompts. The experience should be: Connect Shopify → Products, inventory, policies and orders sync automatically → AI creates its persona → Merchant answers 5 simple A/B scenarios → Deploy Then customers can: Ask → Get recommendations → See product cards → Add to cart → Track orders And merchants can see: Conversation → Cart → Order → Revenue That last part matters. “1,204 AI messages” is an AI metric. “$2,831 AI-assisted revenue” is a business metric. This changed how I think about CAAC Go. We shouldn’t build a better chatbot. We should take Shopify’s commerce primitives to parity, then push the action layer further: Shopify + booking + CRM + WhatsApp/LINE/IG + external APIs + custom actions. Shopify Inbox can become an AI sales associate. CAAC Go should become an AI agent that runs customer operations. Where would you draw the line between matching the incumbent and building differentiation? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

AI isn’t replacing headcount. It’s repricing headcount.

David Sacks @DavidSacks

a month ago

Narrative violation. WSJ today.

284 434 6K 618K 421

@Benioff @salesforce The enterprise AI moat is shifting from intelligence to trust. Agents are easy to demo. Agents you can trust with government-grade access are much harder to build.

I stopped writing “Top 10 Customer Service Tools” articles. LLMs can generate those in seconds. Instead, we’re writing what AI can’t: Original product decisions. Real experiments. Customer failures. Metrics. SEO in the AI era isn’t about producing more content. It’s about publishing information that only your company knows. That’s the content we’re betting on while building CAAC Go. What kind of content do you think survives in an AI-first search world? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

I thought importing a Zendesk Help Center would be a crawling problem. It wasn’t. Zendesk Help Centers sit behind Cloudflare bot protection, so crawlers hit a 403. Ours did. Every vendor’s does. The mistake was treating an integration problem like a scraping problem. So I changed the approach. Now customers paste their Zendesk Help Center URL into CAAC Go, and we pull the content through Zendesk’s public API. Titles. Full articles. Multiple languages. I tested it on our own Help Center: 177 articles imported, and the agent answered product questions correctly. The limitation is simple: the Help Center must be public. For private Help Centers, customers still need to export the articles and upload them as files. The bigger lesson: When the website fights you, stop fighting the website. Find the system behind it. How often are we building around the interface when the real product lives in the API? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

@chamath Bottom-up isn’t dead. It just won’t stop at the individual anymore. The “aha” still happens with one user, but adoption increasingly depends on organizational trust.

@chamath The biggest shift isn’t from SaaS to AI. It’s from software moats to context moats.

The AI Singularity The argument goes like this: 1. Humans build an AGI. 2. The AGI becomes good at AI research. 3. It designs a smarter AI. 4. That smarter AI designs an even smarter AI. 5. The cycle repeats faster and faster. Looking at the results and capabilities from the various labs over the past few weeks I would say we are firmly in this loop now. The next 18months will be wild. Recursive self improvement will dramatically increase capability very quickly from here. Marginal costs of all models will go to ~$0.

blog.cresclab.com/en/ai-receptio… I stopped treating SEO as a content problem. Our early CAAC Go data showed Australia had healthier activation than most markets—but the sample was too small to trust. So instead of publishing another generic “AI customer support” article, we turned the signal into an experiment. We published a guide for Australian businesses searching for an AI receptionist. Not a product announcement. It covers local pricing in AUD, Australian privacy law, voice vs. website agents, and the buying questions owners actually ask. The goal is simple: Content brings Australian traffic. Product data tells us whether that traffic activates. Then we decide whether Australia deserves more budget. SEO should not run beside product experiments. SEO should be one of them. How closely does your content strategy follow actual product data? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

More context makes an AI agent smarter. It can also turn customer support into surveillance. We just built page context into CAAC Go. When a visitor sends a message, we save the page they were on. Then I looked at a more powerful design: recording the visitor’s entire browsing trail before they ever speak. That would tell the agent: They visited pricing three times. Read the security page. Then asked about the enterprise plan. The support value is obvious. So is the privacy cost. For anonymous visitors, “better context” means tracking behavior they never intentionally shared with the agent. This is not a feature decision. It is a trust decision. Especially after promising users that they own their data—and when serving markets like Singapore, where data storage matters. My current principle: Capture the minimum context required to help, not the maximum context technically available. Would you give an AI support agent the full browsing trail, or only the page where the conversation started? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

I thought India was sending us low-quality traffic. The data proved me wrong. India is our largest source of signups. Activation still has room to improve, but it’s performing better than we assumed. Singapore and Australia look healthier after signup, but the samples are still too small to trust. Once we combined activation with acquisition cost, the surprise became clearer: India, Singapore, and Australia are all within roughly 10% of each other on cost per activated user. Meanwhile, Malaysia and the US are underperforming. So we’re changing the plan: Stop spending on Malaysia, the US, and the UK. Keep India running. Move more budget toward Singapore and Australia, then wait for larger samples. Traffic quality is not a country stereotype. It’s a funnel calculation. Without activation data, cheap traffic can look good. Without acquisition cost, high activation can look good. You need both before deciding where to invest. Would you optimize for the proven volume in India, or place a bigger bet on the early signals from Singapore and Australia? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

Cheap traffic can be the most expensive traffic. Most of CAAC Go’s traffic currently comes from India. The ad metrics look fine. CPC looks efficient. But when we looked past acquisition, many users were getting stuck at the first activation step. They signed up, but never trained the AI, tested it, or deployed the widget. That changed how I think about paid acquisition. A low CPC tells you that you bought attention cheaply. It does not tell you whether you reached people with the problem, tools, and intent required to use your product. So our next move isn’t to spend more. We’re adding country-level funnel tracking to see where users move from: Visit → Signup → Setup → First Value Then we’ll compare India with traffic from markets like Australia. Traffic quality should be measured by activation, not clicks. Would you optimize the product for the traffic you have, or shift acquisition toward markets that activate better? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

PLG doesn’t remove sales. It turns sales into software. In sales-led growth, humans fill the gaps. Sales explains the value. CSMs help customers get started. FDEs solve technical problems. Account executives close the deal. In PLG, nobody is coming to help. The landing page must deliver the pitch. Onboarding must perform discovery. Demo data must replace the sales demo. In-product guidance must replace the CSM. The upgrade prompt must close the deal. That changed how I allocate product effort for CAAC Go: 25% Build 30% Activate 15% Package 20% Acquire 10% Monetize Activation gets more investment than the product itself because, in PLG, the product is also the sales team. Visit → Signup → Setup → First Value → Habit → Upgrade → Expansion Every step is both a product problem and a GTM problem. Would you invest more in building features or activating the ones you already have? — Building CAAC Go in public. 🌐 caacgo.com Follow the journey as I build an AI-native customer support SaaS from scratch. #BuildInPublic #AIStartup #SaaS #AIAgents #CustomerSupport #IndieHackers #Founders #AI

My current allocation for building a PLG product looks roughly like this: 25% Build 30% Activate 15% Package 20% Acquire 10% Monetize The biggest difference between PLG and sales-led growth is not that PLG has no go-to-market motion. It’s that PLG has to encode the work of Sales, CSM, and Solutions Engineering directly into the product. In a sales-led motion: Sales explains the value. CSM helps with setup. A solutions engineer handles technical friction. The account team follows up when the customer gets stuck. Sales closes the deal. In PLG, nobody is there to fill those gaps. So the product has to do all of that: The landing page has to deliver the pitch. Onboarding has to perform discovery. Demo data has to replace the sales demo. In-product guidance has to replace the CSM. Templates and automation have to replace implementation support. The upgrade prompt has to close the deal. This is why activation is much closer to the product’s core in PLG than it is in SLG. The real journey isn’t just: Visit → Signup → Paid It’s: Visit → Signup → Setup → First Value → Habit → Upgrade → Expansion Every conversion point is simultaneously a product problem and a GTM problem. AI has made the “Build” part dramatically faster. But that doesn’t mean building a PLG company has become easy. It means more of the difficulty has moved into activation, distribution, packaging, and monetization. — Building CAAC Go in public: an AI agent for your website. 🌐 caacgo.cresclab.com Follow the journey as I build and grow an AI-native SaaS from scratch. #BuildInPublic #PLG #SaaS #AIStartup #AIAgents #ProductLedGrowth #GTM